Post-stroke fatigue: a scoping review

Background: Post-stroke fatigue (PSF) is one of the most common and frustrating outcomes of stroke. It has a high prevalence and it can persist for many years after stroke. PSF itself contributes to a wider range of undesirable outcomes that affect all aspects of daily life. The aim of this review was to identify and summarise the most recent research on PSF, in order to update the evidence base. Methods: We updated an existing review (Hinkle et al. 2017) systematically searching CINAHL, MEDLINE, PsycINFO, and PubMed to cover new research studies between 1 st March 2016 and the search date (19 th January 2020). We included interventional and observational research, and clinical practice guidelines that were not covered in the original review. After duplicate removal in EndNote, two reviewers screened the search results in Rayyan, and data from eligible full texts were extracted onto an Excel spreadsheet. Finally, we used RobotReviewer and a human reviewer to assess the risk of bias of randomised trials for this scoping review. Results: We identified 45 records for 30 studies (14 observational, 10 interventional studies, and 6 guidelines). Apart from one, the interventional studies were single-centred, had high risk of bias and small sample size (median 50). They investigated exercise, pharmacotherapy, psychotherapy, education, and light therapy. Observational studies mainly reported the factors related to PSF including co-morbidities, depression and anxiety, quality of life, activities of daily living, stroke severity, medication use and polypharmacy, polymorphism, pain, apathy, limb heaviness, neuroticism, mobility, and thyroid-stimulating hormone. Guidelines either did not report on PSF or, when reported, their recommendations were supported by little or low level of evidence. Conclusion: Although we identified a number of recent studies which have added to our current knowledge on PSF, none are robust enough to change current clinical practice.


Amendments from Version 1
In Table 2, we replaced the red and green colour keys with grey and white. In Table 4, we added the effect sizes for interventional studies as reported in the original studies. In the Discussion, we added the need for future research to include sensitivity analysis and the need for international standard reporting guidelines for measuring and reporting the outcomes. We suggested that future reviews should consider the impact of research on online delivery of interventions.

Introduction
Post-stroke fatigue (PSF) has been defined as 'overwhelming feeling of exhaustion or tiredness', which is unrelated to exertion, and does not typically improve with rest 1 . It is one of the most common outcomes of stroke and its prevalence varies between 25% and 85%; however, it is generally accepted that it affects 50% of people after stroke 2 . PSF is linked to undesirable stroke outcomes and affects patients' participation in studies, adherence to medication, and effectiveness of rehabilitation 3 . This has a negative impact on patients' quality of life and daily life activities 4-7 , and also contributes to the burden on family members and carers 8 .
Although researchers have attempted to explain PSF mechanisms 9 , its aetiology still remains unclear. This is partly because there are many contributing factors to PSF 8,10-32 , and each research team may focus only on some of the factors to find a route for preventing, treating or managing PSF. Any endeavour to find the most effective intervention in the research literature leads to a collection of heterogeneous interventions from physiotherapy 33 and exercise 34-38 to psychotherapy, pharmacotherapy, and recently laser therapy 39-41 .
As a systematic effort to review these scattered interventions, a Cochrane review 42,43 compared all the tested PSF treatments to a control group, to standard care, or to each other, through reviewing randomised controlled trials (RCTs). This review concluded that there was insufficient evidence of the efficacy of the tested interventions in trials, and more robust research with adequate sample sizes was required 42,43 . Since then, more recent systematic reviews until 2019 have attempted to summarise the evidence of effectiveness of Modafinil, mindfulness training, a traditional Chinese medicine, and smart technologies, but still came to a similar conclusion to that of the Cochrane review in 2015 [44][45][46][47] .
As a result of such uncertainty, current clinical practice guidelines rely on low levels of evidence, such as expert consensus, to make recommendations for PSF 48,49 . However, the efforts to design and test treatments continue, which makes it necessary to keep up-to-date with new research and practice literature.

Objective
The objective of this review was to identify and summarise the most recent research literature related to PSF in order to update the evidence base. As there was an existing review covering the literature up until 2016 50 , we only updated the literature not covered in this review.

Scoping review methods
We followed Arksey and O'Malley framework 51 for conducting this scoping review. We also used Preferred Reporting Items for Systematic Reviews and Meta-Analyses-Extension for Scoping Reviews(PRISMA-ScR) for reporting 52 . The relevant PRSIMA-ScR checklist is available as Extended data and the flow diagram is reported in the Results section ( Figure 1).

Search methods
We ran a search to include studies in the English language only, between 1 st March 2016 and 19 th January 2020 (search date) in CINAHL via EBSCOhost, MEDLINE via Ovid SP, PubMed (excluding MEDLINE), and PsycINFO via Ovid. There were no limitations to document type (e.g. thesis), study completion status (e.g. ongoing), and publication status (e.g. unpublished) at the search stage. We report the search strategies for all databases in Extended data.

Selection of studies
We imported the search results into EndNote X6 and de-duplicated them based on title, and additionally doublechecked the automatically identified duplicates manually. Two reviewers (GA and FS) screened the results independently against the eligibility criteria using Rayyan, which is a recommended screening system 53 . Discrepancies were resolved through discussions or asking a third reviewer (AD).
Two reviewers (GA and FS) also investigated the full texts of relevant search results against the same criteria involving a third reviewer (AD) in case of disagreement. At full text screening stage, we also investigated the reference lists of the relevant studies to identify additional relevant studies. Since one study may have multiple reports or publications, we kept a record and cited all the reports of a single study to provide a better overview of the new research evidence.

Eligibility criteria
We included the following studies: -Studies of adult humans with PSF -any definition of PSF -at any stage of the stroke care continuum; -Any interventional (clinical trial) or observational (cohort, case-control, and cross-sectional) studies, and clinical practice guidelines; -Studies reporting findings that had not been included in the previous review; -Studies included in relevant systematic reviews.
We excluded the following studies: -Ongoing studies or protocols with no results (listed and cited in this paper for further follow-up); -Tool validation studies without reporting new findings on PSF.

Data extraction methods
One reviewer (GA) extracted and entered the data in Excel 2007 and the second reviewer (FS) checked the extracted and  For clinical practice guidelines, we extracted the following data: -Study name and year; -Country and organisation who produced the guideline; -Recommendations on PSF; -Evidence base reporting the level of evidence or study designs related to the level of evidence.

Quality assessment methods
We used RobotReviewer for assessing the risk of bias in the four categories of the Cochrane Risk of Bias tool 54 for included RCTs. Although this automation system is reliable for checking the risk of bias for certain bias categories 55,56 , one of the reviewers (GA) also double-checked and revised RobotReviewer's assessment and corrected the data where necessary. We also added a 'selective reporting of outcomes' category to the list of biases to cover the main biases in Cochrane Risk of Bias tool. Because of 'scoping' nature of this review and lack of time and resources, we did not assess the risk of bias for non-RCTs.

Synthesis methods
We summarised the data from the new relevant literature in tables. We did not proceed to a meta-analysis for fatigue outcomes due to the heterogeneity of studies. We checked if any of the interventional studies considered following the CONSORT 57 for reporting RCTs or TIDieR checklist 58 to report the components of new interventions.

Results
The search identified 1021 results. After screening, we included 45 relevant records related to 24 studies and 6 guidelines ( Figure 1).  71 . In studies that reported the intervention delivery details, the psychological interventions were delivered individually and face-to-face -rather than online -by psychologists. We also assessed the risk of bias for RCTs and reported the categories of risk in Table 2 with supporting statements in Extended data.
Most of the interventional studies have a medium to high risk of bias. Table 2 shows only two studies in white cells (indicating low risk of bias) but both have small sample size consisting of 34 (MIDAS study 63-68 ) and 64 randomised patients 61 respectively.
We identified 14 observational studies of which half had a prospective cohort design 77-90 and the other half were cross-sectional surveys 91-97 . Three cross-sectional surveys were embedded within cohort studies 91,94,97 . Only one of the studies (NotFAST) had a follow-up report [81][82][83][84][85] . Details of all studies are reported in Table 3 as well as the Extended data. The majority of observational studies investigated factors related to PSF including co-morbidities, physical and mental outcomes, illness characteristics, characteristics of interventions, and biomarkers (Table 5).
We identified six recent guidelines from three Englishspeaking countries including the UK 49 and two North American countries (one from Canada 48 and four from the USA 98-101 ). Among these, the Canadian guideline was the most recent and the only one with comprehensive recommendations   *For cohort studies, the left number shows the number of participants who finished follow-up, and the right number is the number of participants who started and took part in the study; for cross-sectional studies within cohort studies, the left number shows the number of participants in cross-sectional study and the right number is the number of participants in cohort study.
** Sometimes reported as time period and sometimes as mean and standard deviation across the studies.
on PSF. The UK guideline will be updated in 2021. Half of these guidelines, that is, all those from USA, have not provided specific recommendations on PSF, as reported in Table 6. In almost all the guidelines, the reliance on 'experts' consensus' is apparent because of the limited evidence base for PSF (Table 6).

Discussion
We Despite the high prevalence of PSF 2 and its obvious effects on treatment adherence 102 , in practice, only half of recent stroke guidelines have clinical recommendations on PSF. Of those that do, two guidelines provide only brief recommendations, and only one provides comprehensive recommendations, but these are based on low levels of evidence 48 . The weak evidence base and the need to rely on expert consensus is *Grey cells contain findings from low risk studies; however they have small sample size. We reported the data as reported in the original report.
likely to be the main reason that PSF is generally not covered in the guidelines. While the observational studies reported the type of stroke, the interventional studies did not include this important data, which makes it difficult to summarise studies. Most of participants entered the interventional studies three-months after stroke. This is likely to be due to a number of reasons; for example, fatigue is not recognised immediately after a stroke, some studies want to ensure that participants have a continuous fatigue, and there is competition for recruitment in the early stages to more acute trials. However, one issue worth considering is whether the construct of PSF holds for fatigue experienced in research participants recruited years after their stroke, and whether this fatigue is a function of other issues. Future systematic reviews could address this issue by conducting sensitivity analyses comparing studies that include participants many years after their stroke with those including participants immediately after their stroke.
The variety of the interventions tested in studies and trials underlines the complexity of PSF and is an indication to researchers that future interventions will probably need to target multiple aspects of fatigue. While current reporting practice of interventions in RCTs included in our review is of concern (none followed TIDieR and two followed CONSORT), future studies should consider following reporting guidelines such as CONSORT and TIDieR for interventional studies, STROBE for observational studies, and RIGHT 105 , AGREE 106 , or CheckUP 107 for clinical practice guidelines. In addition, harmonisaton of studies requires standard international guidelines regarding outcome measurements and time points for measuring PSF in a standard way to create a homogenous and collective body of evidence.
Among the observational studies, the population-based study from the stroke register in New Zealand 91 and Sweden 77 provides valuable insights about the link between co-morbidities and increased PSF in long-term (4-7 years). This, and other similar register-based studies, represent the added value of having high-quality data in health system databases for long-term observational and register-based studies 108 .
Psychologists delivered the psychotherapies in RCTs to individual patients and there was no intervention using online platforms as the media of delivery. This may be due to a number of reasons: it is usual to test the efficacy of an intervention face to face before moving to another medium; participants with stroke may have other problems which mean it is more difficult to deliver treatments online, e.g. communication issue and cognitive problems. Online delivery of such interventions is becoming more common in some clinical services, and new research is emerging 109 . Therefore, future reviews may wish to consider the impact of such interventions delivered online.
Fatigue Severity Scale (FSS) was the main outcome measure for PSF in observational studies, whereas Fatigue Assessment Scale (FAS) was used more frequently than other measures in interventional studies. Bearing in mind that both these PSF measurement scales are valid and reliable, the main reason that the FSS has been used more frequently is probably because it is now seen as a way to compare different studies: in simple terms, researchers use it because other researchers have used it. It is also relatively straight forward to complete.
Only one of the observational studies and half of the interventional studies were registered in clinical trial registers, with the remaining unregistered trials potentially introducing bias in selective reporting of outcomes 110,111 . One of the interventional studies was registered retrospectively with potential for the same bias 69,70 .

Limitations and strengths
It is possible that we overlooked studies that did not report PSF in the searchable part of the paper or if the report was not indexed in the searched databases. In such cases, we invite readers of this review to contact us or comment on the paper online.
Due to the challenges we faced in identifying the search strategies in the previous review when conducting this update, we have explicitly documented our search criteria and strategies so that this review can be easily repeated and updated by future researchers. We feel this is a key strength of this review. We also feel that the use of automation tools such as Rayyan and RobotReviewer for this evidence synthesis was a strength in terms of saving time and other resources while still maintaining the quality of the review. Finally involving a multi-disciplinary team of clinicians and methodologists (information specialist and systematic reviewer) allowed us to consider both the clinical and methodological aspects of the studies in this review.

Conclusion
The current trend of research on PSF shows the continued importance of this topic globally. Our review identified a weak evidence base that highlights the need for more research

Open Peer Review
The current review updates the work of Hinkle et al. 2016, and the rationale for doing so is clearly stated by the authorship team. That the review here generated a further 24 studies (and six sets of guidelines including PSF) since 2016 further demonstrates the growing impetus of work in this area. The objective stated is rather non-specific, but this is acceptable in such a scoping review that serves as a summary of recent evidence. It is unsurprising that this review concludes that further research in this area is required, and the authors make relevant suggestions as to the characteristics of future research that are clearly based on the findings of this review.
The review is clearly written, and the methods used and ensuing findings have been reported with transparency and considerable attention to detail. Search strategies are available via an embedded link. This paper makes a very useful contribution and provides a foundation for further work.
The following minor comments and suggestions may be of use to the authors: The use of the word 'following' in the opening two section headings of the methods might be worth reconsidering, readability of this phrase in a title is challenging (page 3).

○
In Figure 1, suggest clarifying '34 not new/covered'-does this mean 34 records removed as covered in the previous review? (page 4).

○
Section on data extracted from clinical practice guidelines, please clarify 'study name and year'-does this refer to the name and year of the guidelines or the included studies from which the guidelines was written? (page 5).
○ Was any specific tool used to assess risk of bias in the non-randomised studies? If so, this should be stated in the methods text. If not, some justification would be helpful.

○
In table 3, the final column is not entirely clear-is this mean time after stroke onset for included participants? Simple clarification in the column heading or legend would address this (page 7/8).
○ Suggest rephrasing 'probably the most effective interventions' to deliver a clearer message here (page 9).

○
In the discussion, the section on online platforms is somewhat unexpected-possible delivery via online mechanisms does not seem to have arisen prior to this point, though I apologise if I have missed this. Perhaps a line to place this paragraph in context might help the interpretation here? (page 10).
○ As the authors focus on possible reasons for the use of the Fatigue Severity Score (FSS) in one section of the discussion, it might be helpful to include a line about the validity/reliability of this measure at this point.

○
There are a few very minor typographical and grammatical errors.
○ Are the rationale for, and objectives of, the Systematic Review clearly stated? Yes training to two PhD students who were involved in this review, we added risk of bias assessment only for the RCTs. We did not assess non-RCTs for risk of bias because our resources were limited. We have now addressed this in the text: "Because of the 'scoping' nature of this review and lack of time and resources, we did not assess the risk of bias for non-RCTs." Department of Clinical Neuroscience, Rehabilitation Medicine, Institute of Neuroscience and Physiology, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden I found this scoping review very exciting to read since it addresses a very important issue that needs to be managed daily in post-stroke rehabilitation. I believe that the interest in post-stroke fatigue is great from both researchers and clinically active colleagues. I am positive to and welcome this type of "game-changer" by a transparent scientific publishing.
The title identifies the manuscript as a scoping review and reflects the population (individuals post-stroke) and the construct (post-stroke fatigue). The abstract is presented with a clear structure, including background, methods, results and conclusion, of which all parts relate to the aim of the study. The rational for the review is described; there is insufficient evidence of the efficacy of the interventions related to post-stroke fatigue in conducted trials and more robust research with adequate sample sizes is required. In addition, it was time to make an update. The objective is well described, to identify and summarise the most recent research related to poststroke fatigue in order to update the evidence base. The update covers new research studies between 1 st March 2016 and 19 th January 2020. The search strategies for the databases are reported in a separate link (Extended data). In the results, data related to the objective of the study is presented with a clear structure in six tables. These tables include characteristics of, and summary of the findings from, the included interventional and observational studies, risk of bias and a summary of included guidelines. Due to the heterogeneity of studies, no meta-analysis for fatigue outcomes was performed. The interventional studies were limited by either having high risk of bias or small sample size. Consequently, transferring the research findings to practice was hard.
The current manuscript is well written and easy to follow. It is clear that you have put in a lot of effort and made a great job. Even so, I have following minor comments/suggestions that could possibly improve your article:

Method:
As written, I am uncertain about limitations to language. According to the manuscript, there were no limitations to language. However, in the Extended data, English language seems to be a limitation in the search strategies. Please, clarify so there is no doubt about this. ○ Figure 1: Please, consider clarifying the number of excluded reports per each specified criterion regarding the 473 excluded reports; no clarification (in numbers) has been made for them as for the 86-excluded full-text reports. ○ Figure 1: Please, give the readers more information about the nine additional sources. Which was your strategy, how did you found them? Did all arrived from the other articles´ reference lists? ○

Results:
Are MIDAS (presented in Table 1), and ARCOS-IV and LAS-1 (presented in Table 3) examples of studies not yet published? ○ (In Table 1, the fourth column, there is space available (two rows are already used) for you to write "Number of participants" or "No." instead of using the symbol "#".) Table 2: I like the use of different colours, and red and green are instantaneous to understand. However, in color blindness, the most common difficulty is to distinguish between red and green. Perhaps you can choose another colour combination. In the online version of Robot Reviewer report, I think that the table Risk of bias has a more easy to read layout than Table 2 in the main manuscript. ○ (Table 3: Consider using "Number" or "Number of" or "No." or "No. of" instead of "#" in the third and the fifth column?) ○ Table 3, sixth column, "Stroke type": Please, review the use of "/". Should you use "and", "," or delete the "/" somewhere? Regarding Kuppuswamy and NotFAST: "First", but which stroke type?
○ (Table 4, fourth column: I suggest you to use "Post-stroke fatigue finding" instead of the use of the abbreviation "PSF finding". Avoiding unnecessary abbreviations makes reading easier.) ○ (In Table 5, consider if you should specify "Finding" to "Post-stroke fatigue finding" in the light blue heading (in line with the selected sub-heading in Table 4).)

○
In the data commentary related to Table 6, you state that the Canadian guideline was the only one with comprehensive recommendations on post-stroke fatigue. In Table 6, under the heading "Recommendation" the reader is referred to page 15-16 of the guideline. I suggest that you present vital parts of the content in Table 6, in addition to this reference. I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 23 Jun 2020 Ghazaleh Aali, University of Nottingham, Nottingham, UK We are grateful to the expert reviewer for their interest in the topic and for spending time in reviewing this manuscript. We value their positive and thorough comments and have revised the manuscript accordingly.

Method
Comment: As written, I am uncertain about limitations to language. According to the manuscript, there were no limitations to language. However, in the extended data, English language seems to be a limitation in the search strategies. Please, clarify so there is no doubt about this. Reply: Thank you for this comment. We initially intended to have no language restrictions but, as we followed the methods from an existing scoping review, we needed to apply a language limitation. This has now been corrected to: "We ran a search to include studies in English language only…".
Comment: Figure 1: Please, consider clarifying the number of excluded reports per each specified criterion regarding the 473 excluded reports; no clarification (in numbers) has been made for them as for the 86-excluded full-text reports. Reply: We followed the PRISMA reported guideline and flow diagram which does not specify recording or reporting detailed reasons for exclusion at the Title/Abstract Screening step (Liberati et al. 2009; DOI 10.1136/bmj.b2700); however, as it is mandatory to report the reasons for exclusion in the Full Text Screening step, we have done this.
Comment: Figure 1: Please, give the readers more information about the nine additional sources. Which was your strategy, how did you found them? Did all arrive from the other articles' reference lists? Reply: We have now included this statement under the Selection of studies section: "we also investigated the reference lists of the relevant studies to identify additional relevant studies ".

Results
Comment: Are MIDAS (presented in Table 1), and ARCOS-IV and LAS-1 (presented in Table  3) examples of studies not yet published? Reply: No. These studies were completed and have reported their results, so they met our eligibility criteria. If a study was 'ongoing' at the time of our review, we did not report them in the table because "Ongoing studies or protocols with no results"was one of our exclusion criteria. If a study had no specific name, we used the last name of the first author and the year of publication as the study name in the tables.

Comment:
In Table 1, the fourth column, there is space available (two rows are already used) for you to write "Number of participants" or "No." instead of using the symbol "#". Reply:We have corrected this now to "No. of".
Comment: Table 2: I like the use of different colours, and red and green are instantaneous to understand. However, in colour blindness, the most common difficulty is to distinguish between red and green. Perhaps you can choose another colour combination. In the online version of Robot Reviewer report, I think that the table Risk of bias has a more easy to read layout than Table 2 in the main manuscript. Reply: We followed Cochrane's Risk of Bias tool and therefore also followed their reporting method. However we recognise that some of our readers might be colour-blind and so we are using plus signs and question marks in addition to the colour coding. We have also now changed the colours from red/green to grey/white. Table 3: Consider using "Number" or "Number of" or "No." or "No. of" instead of "#" in the third and the fifth column? Reply:We have corrected these to "No. of". Table 3, sixth column, "Stroke type": Please, review the use of "/". Should you use "and", "," or delete the "/" somewhere? Regarding Kuppuswamy and NotFAST: "First", but which stroke type? Reply: We replaced "/" with" ,". We used type of stroke as reported in the studies. Thus, although 'First' does not refer to a specific stroke type, this was what was reported and therefore what we presented under 'type'. Since this is a scoping review, we did not contact specific researchers to clarify these details. Table 4, fourth column: I suggest you to use "Post-stroke fatigue finding" instead of the use of the abbreviation "PSF finding". Avoiding unnecessary abbreviations makes reading easier.

Reply:
We have now corrected this to 'post-stroke fatigue'. Table 5, consider if you should specify "Finding" to "Post-stroke fatigue finding" in the light blue heading (in line with the selected sub-heading in Table 4).

Reply:
We have now added 'post-stroke fatigue'.

Comment:
International recommendations regarding outcome measures related to, and time points for measuring, post-stroke fatigue in research studies on recovery after stroke would probably reduce the heterogeneity of studies and facilitate further summaries and updates. You could possibly include that post-stroke fatigue might be an issue for future Stroke Recovery and Rehabilitation Roundtable (SRRR) work.

Reply:
Thank you for this suggestion. We have made a comment about this being an important area for future research but do not want to identify an individual group or organization to take this forward. We added: "In addition, harmonisaton of studies requires standard international guidelines regarding outcome measurements and time points for measuring PSF in a standard way to create homogenous and collective body of evidence.".

Comment:
You have a separate paragraph regarding limitations. Have you thought about including a sentence related to strengths of your study, in a paragraph in close proximity to "Limitations"? Reply:Thank you for this comment. We have changed the heading to 'Limitations and Strengths' and added three strengths of the review in this section: A. Public and open sharing of our methods so that anyone can update our review, B. Utilizing automation tools such as Rayyan and RobotReviewer to save time and resources and C. The fact that this was A multi-disciplinary review team.